AI Agents in Customer Scenarios: How I Check Whether a Client Holds a Financial Licence

For one task it matters to me whether a user is an employee of a regulated organisation: does their employer hold a financial licence?

How does this work in the pre-AI era? You put a form on the site: company name, website, licence, regulator. Then you ask them to attach the licence and the standard set of documents.

You hire somebody. They find the regulator's website, check the entries on the licence against the register, and confirm the licence is current. The regulator rarely publishes the official website, so what you have is a phone number, and from there you can do whatever you like with it: practise your French, for example.

Any internet marketer will tell you the basic rule - the more seamless the customer experience, the better the chances they do not get lost on the way to conversion. If a trial means begging the accounts department for a copy of the director's passport, then you had better be building some very magical product. For a company with fifty years of history that may be normal, but a startup will not survive it.

I reduce the task to this:

1. Validate the email address: does the applicant really have access to it?
2. Find the organisation's official website and compare it with the domain of the applicant's email.
3. Find the licence details on the organisation's website.
4. Find out who the official regulator is for this type of organisation in this country.
5. Check that the licence is valid on the regulator's website. Compare everything the regulator gives: address, domain and so on.
6. Check the secondary signals: domain age, the applicant's IP, previous experience with this organisation and this subnet, global watchlists and so on.
7. Ask for whatever was missing.

The "site" calls a claude code instance right there on the same server, and it runs this check autonomously and scores the user. If something looks off, it goes to manual review. It hardly needs saying that the first version of this automation takes less time to build than it took me to write this post.

After that it all depends on how deep the due diligence has to go, and on your imagination: it can call them as well, in the language of the country.

Where this can be used

• Trials. Approve it, or does the company profile not look like a target customer?
• Service priority. Somebody pressed contact or demo: do they get the founder's calendar, a regular salesperson's, or a polite brush-off?
• Pricing. Would rather not embarrass wealthy enterprise customers with a cheap price? Just bear in mind that this is not legal everywhere.
• Pre-checking documents. Upload your financial statements and I will tell you whether we can lend to you (sensitive data aside).
• Enriching a CRM. Usually there is no need to do this in real time, but the principle is the same. Although here access to proprietary data will usually be worth more than poking around a customer's website.